The looming on-device AI era could bring along a familiar headache
Echoes of the bring your own decide tsunami are showing up as models move closer to the edge.
Large companies already have a laundry list of hang-ups and headaches in the way of getting modern language models actually into production. But if one of the dream scenarios of AI development plays out as it is set to, there will be a whole new realm of challenges that echo the emergence of the mobile era.
While foundation model providers chase creating powerful one-size(ish)-fits-all models, startups like Mistral and oLlama are quietly pushing development of language models run entirely on local devices. And if the current development arc holds, enterprises may end up reckoning with employees bringing—or even deploying—a whole universe of new unsanctioned products intermingling with their work.
If that sounds familiar, it should: enterprises had to manage a very similar workplace reset with the emergence of the “bring your own device” (or BYOD) paradigm. Whether enterprises wanted to or not, employees craved to use their personal devices—technically superior consumer-oriented products—to do their work.
That shift wreaked cultural havoc on enterprises scrambling to keep their data and operations behind firewalls. Even compromises like issuing BlackBerry phones, which were almost exclusively pitched as enterprise-grade devices, weren’t enough. Inevitably companies caved and started to build their workflows around personal devices, rather than end up as adversaries to their own employees. And it provided the benefit of enabling employees to work on the go without a prohibitive level of investment on the part of the company.
The advent of oLlama and tooling in its immediate proximity (more specifically, Llama.cpp) that are pushing powerful custom models toward the edge could force companies to come to a similar moment of reckoning: do we allow the usage of external AI products, and if so, to what extent do we try to control it?
“There’s always been a tension between companies wanting to clamp down on everything internally and employees demanding better experiences and products,” Adam Wenchel, CEO of AI security startup Arthur, told me. “It’s going to be a question of how aggressive enterprises want to be. You have to either block the app out and risk annoying employees, accept that there’s some risk, or have an enterprise-enabled version available.”
Echoes of the 2010 mobile revolution, once again
While Steve Jobs was igniting the mobile revolution with the release of the iPhone 4 and iPad, Apple’s now-CEO Tim Cook and other executives worked furiously behind the scenes to woo enterprises to cave to the inevitable. The launch of the iPhone 4 and iOS 4 in 2010, which introduced better mobile device management (MDM) features, essentially sealed the deal, leading to two industry-altering developments:
Most enterprises caved, allowing employees to use their own phones and tablets for work purposes, under the conditions of using MDM and VPNs when accessing any proprietary information.